Thanks all for the replies.
I have FW build 1477971088.
I tried it again with a regular 10k resistor (9.85k on the Fluke) and on the 100k range it floated slowly from an initial 14k to about 7k when I gave up. A 220k on the 1M range, drifted down slowly from 250k to 180k.
So a slightly inconsistent result, but indicating that something is not OK about the current source or the way of reading voltage.
@Tim no the thermistor and the discrete resistors are all loose parts, not on a PCB.
I wondered about leakage from contamination on the PCB, but this would probably affect all ranges equally, and in my earlier tests, the 10k range seemed to work well enough.
I’ve had long discussions with James in the past, about the real power measurements which came after some development work.
I’ll try an upgrade and see what I see.